Professional Documents
Culture Documents
COIMBRA — 2005
1
Demand for Tourism in Portugal: A Panel Data Approach
Sara A. Proença
(sproenca@esac.pt)
Escola Superior Agrária, Instituto Politécnico de Coimbra
Elias Soukiazis
(elias@fe.uc.pt)
Faculdade de Economia, Universidade de Coimbra
Abstract
The tourism activity in Portugal is responsible for about 8% of the national product and
employs 10% of the total labour force. In addition, the receipts from tourism contribute
substantially in financing the current account deficit of the balance of payments in this
country. These are convincing arguments to justify the analysis of the determinants of
the demand for tourism in Portugal and might constitute an important guide-line for the
policy making institutions.
Empirical studies on this field for Portugal have shown little attention in
modelling appropriately the demand function for tourism and identifying the main
sources of tourism flows. The majority of studies take into account the demand side
determinants of tourism, usually proxied by income and price measurements, and little
attention has been given to the supply factors which might influence substantially the
tourism performance. Factors such as infrastructures in networks and accommodation
capacity in the hosting country have been ignored in such studies. Furthermore, the
empirical analysis is mainly concentrated in time-series estimations for separate
countries being potential suppliers of tourism and panel data estimations are
exceptional.
The purpose of this study is to bring into account the above elements to which
little attention has been given in the empirical literature. First, we use a combination of
time series and cross sectional data (panel data techniques) to estimate the demand
function of tourism in Portugal by considering four main countries as the basic tourism
suppliers, Spain, Germany, France and the U.K., responsible for almost 90% of the total
tourism inflows in this country. Second, in the demand function we introduce both the
demand factors (per capita income, relative prices) and the supply factors (public
investment ratio, accommodation capacity) to explain tourism performance in Portugal.
Third, dynamic panel data techniques are used to estimate the demand function of
tourism, avoiding misspecification errors, estimation inconsistency and overall
explaining the adjustment process of tourism flows.
Our empirical analysis shows that per capita income is the most important demand
determinant and accommodation capacity the most important supply determinant
explaining thus the tourism movement in Portugal. The dynamic panel data estimation
highlights the importance of the accommodation capacity as the most important factor
in attracting more tourism to Portugal. It is also shown that the adjustment process is
slow and given the status quo it is difficult to attract more tourism flows at the above
mention countries.
Keywords: Tourism demand, supply and demand determinants, panel data estimation,
adjustment mechanism.
JEL Codes: C23, D12, L83.
2
1. Introduction
Tourism is a sector that involves a multiplicity of economic activities responding to
differentiated demands with specific characteristics at the national and international
levels. The complexity and interaction of the tourism activities justify its consideration
as a special sector that integrates a set of economic activities related mainly to travelling
and accommodation services. The combination of demand (travel decision) and supply
(accommodation provision) characteristics at the national and international levels
creates some difficulties in modelling the tourism activity as a whole. However, the
increasing importance of the tourism sector in terms of its contribution to the national
product, the employment and the balance of payments creates the need to investigate the
determinants of tourism flows within a specific country, and especially for countries
with great dependence on this sector, as Portugal.
The tourism activity in Portugal generates about 8% of the gross domestic
product, represents 10% of the total employment and contributes decisively to attenuate
the current account deficit of the balance of payments. The substantial contribution of
tourism in the Portuguese economy justifies the interest in explaining the determinants
of tourism demand and, therefore, the factors which influence the decision of tourists to
choose this country as a destination place. A better knowledge of the factors that explain
the tourist’s preferences to choose Portugal as a destination place will help the policy
makers to design more adequate strategies in order to develop farther this sector.
Although the importance of tourism in the Portuguese economy is widely
recognised, little attention has been given to explain systematically its determinants.
Empirical studies in explaining the international demand of tourism in Portugal are
limited and the majority of these studies only consider demand factors (personal income
and relative prices) as the main explanatory variables of the tourism demand. The
supply factors have been systematically ignored when the demand for tourism equation
is estimated. Factors, such as, infrastructure networks and accommodation capacity
have not been considered as potential arguments in attracting more tourism inflows. The
econometric models mainly used consider cross-section or time series data and rarely a
panel data approach.
The purpose of this study is to provide an empirical analysis that contemplates the
weaknesses that have been observed with regard the demand function of tourism in
Portugal. More specifically, we introduce into the demand function supply factors as
well and we adopt a panel data estimation approach which allows for specific country
3
effects. A dynamic estimation approach is also used to capture the long term tendencies
of tourism movements.
The remaining of the paper is organized as follows. Section 2 provides a review of
the literature on the demand of tourism explaining the theoretical and empirical aspects.
Section 3 explains the specification of the demand function of tourism to estimate and
analyses the data. Section 4 presents the results from the panel estimations of the
demand function of tourism and discusses policy implication issues. Section 5 estimates
the dynamic demand function of tourism and explains the adjustment process of the
tourism inflows. The final section concludes.
4
the visitors spending; or the stay-length of tourists visiting a country. None of these
measures is fully satisfactory in encompassing all the aspects which characterize the
demand for tourism in a specific location.
González and Moral (1995), in a study about the international tourism demand in
Spain, refer that one of the main problems in analysing the potentialities of the tourist
sector is to find a precise indicator to measure the external demand. Bearing in mind
that the demand for tourism is a variable not directly observable it is necessary to find a
suitable proxy to represent it. The authors use tourists spending as the dependent
variable, defined as the product of three factors: the number of tourists, the length of
their stay and the daily average spending. This is a more complete definition than using
the number of entrances to express tourism demand. The latter does not take into
account the stay - duration and spending behaviour. Cunha (2001) also argues that the
number of entrances is not a good approximation to express tourism demand since it
ignores one of the most important aspects in this sector: the demand of goods and
services that tourists require during their permanence.
Mello and Sinclair (2002), alternatively, use the share of tourism spending of the
origin country to other destination countries to study tourism demand in the U.K. The
authors argue that this variable captures the consumption behaviour of the tourists and
explains the spending component of this economic activity. It is possible to observe an
increase in the tourism inflow accompanied by a reduction in spending explained by
higher domestic inflation and shorter length of stay. For this reason the expenditure
approach is preferable to the inflows approach to study the demand for tourism
behaviour from the point of view of the hosting country.
Rodriguez and Ibanez (2001) use the number of visitors lodged in the destination
country as the dependent variable to study the demand for tourism in a panel data
approach. The choice of this variable to express tourism demand (in comparison with
the number of tourist entrances) has the advantage to consider the length of the stay and
to exclude tourists that are hosted to family or friends houses.
According to the literature review, the most appropriate variable to be used as
the dependent variable in the demand for tourism equation is tourism receipts from the
point of view of the receiving country or tourism spending from the point of view of the
supplying country (Tse, 1999; Lathiras and Siriopoulos, 1998). However, according to
Crouch and Shaw (1992), almost 70% of the studies that estimated tourism demand
functions have used the number of visitors (entrances) as the dependent variable (Qui
5
and Zhand, 1995; Morris, Wilson and Bakalis, 1995; Kulendran, 1996; Akis, 1998). The
main reason for this choice has been the unavailability of data on tourism spending.
6
per capita income as the most appropriate indicator to measure peoples living standards
of the sending country.
According to Witt and Witt (1992) tourism is a luxury good with an expected
income elasticity of demand higher than one and this is what normally occurred in most
studies. On the other hand, Crouch (1995), in an attempt to review the empirical
findings on this subject concludes that income elasticities of the demand for tourism are
specific to each country and no generalization can be made about its value.
1
The substitution effect can be twofold: households can choose to travel inside or outside the country
they live or to choose between alternative places of destinations.
7
1998; Lathiras and Siriopoulos, 1998) introduce the same ratio between different
competing destinations to count for the price substitution effect.
- Other Factors
In some studies, total population of the sending country is used as an explanatory
variable in the demand for tourism function to count for the market size. The rationale
behind this variable is that large countries constitute a potential market for supplying
tourists and, therefore, more economies to scale can be explored.
A trend variable is also used to capture specific households behaviour, such as,
inertia, consumers preferences and habits in this sector. The same variable can also
capture cyclical effects, demographic changes in the sending country or supply
improvements in the receiving country.
The lagged dependent variable in the tourism demand function is normally
included for two reasons: first, to introduce dynamics into the demand function and
second, to capture persistence effects of the tourists behaviour. In general, tourists are
adverse to risk, preferring to spend holidays in places that are already familiar to them
or they had heard something positive about the places they plan to visit (Sinclair and
Stabler, 1997).
Witt and Witt (1995) give another possible explanation for the inclusion of an
autoregressive term in the demand function of tourism: a certain rigidity from the
supply side behaviour. Supply factors related to accommodation capacity, transportation
facilities, human capital qualifications and generally the provision of efficient services
are long term issues involving structural changes and better reallocation of resources in
the sector. Long term or medium term contracts of the operating agencies can be
another source of rigidities as Carraro and Manente (1994) point out.
8
There is a wide variety of model estimation techniques applied to the demand
function of tourism. Using time series data the problem of no stationary data was
recognized, cointegration analysis and ECM estimation techniques were used to ensure
long term properties and non spurious causality between the relevant variables. The
estimated techniques range from ARIMA and Holt-Winters univariate Modelling (Kim,
1999) to 2SLS and 3SLS (Tse, 1999) and ECM estimation models (Kulendran and
Wilson, 2000; Lathiras and Siriopoulos, 1998). On the other hand, panel data
estimations are relatively rare in the empirical literature, especially involving dynamics.
We want to apply this recent technique to the Portuguese case were empirical studies
have not explored yet this alternative methodology.
9
Accordingly, the estimated demand function for tourism in Portugal involves the
following variables:
- Explanatory Variables
1) Demand Factors
As we explained in the theoretical section, the most important factor influencing the
decision of households to travel abroad is their real personal income. As a measure of
the households wealth we use real per capita income of the sending country defined by
the following ratio:
GDPi ,t
Yi ,t = (2)
CPI i ,t ⋅ POPi ,t
where, GDP, POP and CPI are Gross Domestic Product, Total Population and
Consumer Price Index of the sending country, respectively. The source of the data is
OECD (2003), National Accounts.
A second important determinant of the demand for tourism is relative price
between the receiving and the sending countries. Relative price is given by the ratio of
the price index level of the receiving country (Portugal) and the sending country
adjusted by the bilateral exchange rate:
CPI P ,t
Pi ,t = (3)
CPI i ,t ⋅ EX i ,t
where, CPIP and CPIi are the Consumer Price Indexes in Portugal and the sending
country, respectively; and EXi is the real effective exchange rate of the sending country
10
with respect to Portugal. The source of the Portuguese data is INE and OECD for the
other variables.
2) Supply Factors
Supply conditions from the point of view of the hosting country are important factors in
attracting more tourism inflows. Having into consideration the availability of the data,
we introduce two main supply measures.
The first is accommodation capacity (A) measured by the number of beds
available each year to host the tourists who visit Portugal. The data are collected from
INE, Tourism Statistics, several years.
The second is a more general supply measure related to infrastructures (airports,
roads, railways, hospitals, telecommunications, among others) which we believe may
have welfare effects on the daily live of the tourists that visit Portugal. The ratio of
public investment to GDP (IP) is used as a proxy to capture the welfare effects
emanated from public infrastructure networks. The data for the public investment ratio
in Portugal is collected from the OECD (2003), National Accounts.
Finally, a dummy variable (D86) is used to capture the effects of the Portuguese
integration in the EU. The dummy variable takes the value of one in the years followed
the accession to the EEC (since 1986) and zero in the years before the accession. The
idea is to cheque if the border openness with the accession of Portugal in the EEC
provoked a higher inflow of tourists into the country.
Having defined the variables to include in the model we are now able to present
the full specification of the demand function of tourism in Portugal in a log linear form:
ln wi ,t = α i + β1 ln Yi ,t + β 2 ln Pi ,t + β 3 ln At + β 4 ln IPt + β 5 D86 t + u i ,t (4)
11
The data are organized in a panel form with i = 4 and t = 25 giving a total of 200
observations.
2
The estimations have been made by using the econometric program RATS.
12
Table 1. Estimation of the Demand Function of Tourism in Portugal, 1977-2001
(1) (2) (3)
Pooled Fixed Effects Random Effects
Explanatory Variables (OLS) (LSDV) (GLS)
Portugal have higher standards of living, so what matters more in their decision to travel
is their personal income and not the relative cost of living.
13
Public investment ratio in the hosting country has not any significance in the
demand for tourism in Portugal and carries a wrong negative sign. This shows that
tourists care more about their personal accommodation facilities than the welfare gains
derived from public services. On the other hand, the dummy variable introduced to
capture the consequences of border openness after the accession of Portugal in the EEC
shows also no statistical significance but carries the expected positive sign. This is
evidence that the four main clients (Spain, Germany, France and the U.K.) are
traditionally visiting Portugal for holidays and that the integration of Portugal to the EU
did not change their attitude.
Finally, the statistical significance of the individual dummies and the almost equal
value of their coefficients show that differences between the four sending countries are
significant but influence in a similar way the demand for tourism in Portugal. Therefore,
a common constant term can be accepted in the estimated demand equation.
3
The partial adjustment principle admits the following hypothesis: (lwt − lwt −1 ) = δ (lwt* − lwt −1 )
which states that the actual variation of the dependent variable (lwt − lwt −1 ) is a fraction of the desired
variation (lw *
t )
− lwt −1 with lwt* the desired level (not observable) and 0 < δ < 1 the partial adjustment
coefficient, revealing the speed of adjustment: δ → 0 implies lwt → lwt −1 a slow speed of adjustment
(or stagnation) and δ → 1 implies lwt → lw* t a high speed of adjustment or instantaneous adjustment
within the same period.
14
where, the error term, ui ,t , is decomposed into two elements: an individual-specific
However, with this new dynamic specification (5) we face a statistical problem4:
the correlation between the lagged dependent variable and the error term ε i ,
and random effect (GLS) estimations would not be appropriate since the obtained
estimates would be biased and inconsistent, specially in small samples estimations. One
way to solve this problem is to use instrumental variables (IV) estimation techniques.
Then the problem is to find suitable instruments which are highly correlated with the
endogenous regressor but uncorrelated with the individual-specific error term.
Doornik, Arellano and Bond (2002) suggested an alternative method to estimate
dynamic panel data models based on the Generalised Method of Moment estimation
(GMM) which is a generalisation of the (IV) method of estimation. The idea is to
estimate equation (5) by first-differencing all variables (to remove the individual effect)
and then use as instruments all the lagged variables used in the model. Doornik et al.,
(2002) developed a Dynamic Panel Data (DPD) estimation approach in Ox5 to estimate
dynamic panel data models by using the GMM method.
The dynamic specification of the demand function of tourism in Portugal,
equation (5), has been estimated by the GMM method by using orthogonal deviations in
the variables6, as has been suggested by Arellano and Bover (1995). The instrumental
variables used in the estimation were all the predetermined variables, two lag periods of
the variables A and IP , the individual country dummies and the dummy D86. The
estimated results obtained are the following:
4
For more technical details see Greene (2000), chapter 14.
5
The programming procedures to estimate dynamic panel models are available in www.nuff.ox.ac.uk/
Users/Doornik/.
6
The orthogonal transformation of the variables consists in defining the difference of each value in
relation to the average of future values (at each moment) in a way to remove the individual effect without
compromising the orthogonally of the transformed terms of the disturbances, i.e.,
1
⎛ xi (t +1) + ... + xiT ⎞⎛ T − t ⎞ 2
x = ⎜⎜ xit −
*
⎟⎟⎜ ⎟ , t = 1,..., T − 1.
T −t ⎠⎝ T − t + 1 ⎠
it
⎝
7
The Sargan test allows to test the validity of the instruments used in the estimation (accepting the null
hypothesis means the use of valid instruments).
15
ln wi ,t = α i + 0.091 ln Yi ,t − 0.121 ln Pi ,t + 0.920 ln At − 0.425 ln IPt − 0.003 D86 t + 0.780 ln wi ,t −1
t − Stat. (0.103) (−0.530) (2.06) (−1.45) (−0.129) (11.0)
S tan dard Error = 0.004, RSS = 0.341, T ⋅ N = 88, S arg an Test 7 = 50.03 [1.000]
In this dynamic estimation the most significant variable is the lagged dependent
variable, suggesting strong adjustment dynamics in the behaviour of tourists coming to
Portugal. The short run income elasticity of demand for tourism looses its significance
but the accommodation capacity preserves its statistical significance. Therefore, the
dynamic estimation of the demand function of tourism in Portugal suggests that
accommodation facilities is the most important supply factor influencing the decision of
tourists to choose Portugal as the destination place.
The value of the adjustment coefficient8 ( δ = 22% ) gives evidence of a rather low
adjustment process between the actual variation of the demand for tourism and the
desired long-run level. This means that the number of tourists visiting Portugal each
year does not differ substantially from the previous years giving evidence of some kind
of inertia or rigidity in the tourism inflows. This is an expected result since tourism
demand in Portugal is concentrated mainly in four European countries (Spain, Germany,
France and the U.K.). The attraction of more tourists from these countries is a hard task
without improving the quality of the accommodation capacity. Portugal has to develop
new policies to reduce its relative dependence from these four European countries and
to explore new markets characterized by higher standards of living.
5. Concluding Remarks
The basis aim of this study was to estimate the demand function of tourism in Portugal
with respect to four main tourism suppliers (Spain, Germany, France and the U.K.) that
count for 90% of the total tourism inflows in this country. A panel data estimation
approach was used to identify the main determinants of the tourism demand in Portugal
over the period 1977-2001.
The demand for tourism equation was specified in a way to include both demand
and supply factors as the potential determinants in explaining tourism inflows in
8
From the partial adjustment mechanism (lwi,t-lwi,t-1)=δ(lw*i,t – lwi,t-1) we get lwi,t = δlw*i,t + (1-δ)lwi,t-1
and substituting the long-run equilibrium relation lw*i,t = b0 + b1lyi,t + b2lPi,t + b3lAt + b4lIPt + b5D86 we
derive the short-run model lwi,t = δb0 + δb1lyi,t + δb2lPi,t + δb3lAt + δb4lIPt + δb5D86 + (1-δ)lwi,t-1. The
coefficient of adjustment δ is taken from the estimated coefficient of the lagged dependent variable lwi,t-1.
16
Portugal. The panel data estimation approach by using a static specification provides
evidence that per capita income is the most significant explanatory factor from the
demand determinants and accommodation capacity the most significant factor from the
supply determinants. The income elasticity of the demand for tourism higher than one
confirms the usual finding that tourism is a luxury good. The relative cost of living
between the receiving and the sending country and public investment ratio in the
hosting country do not influence significantly the decision of tourists to choose Portugal
as a place of holidays destination. On the other hand, the border openness due to
accession of Portugal in the EEC does not seem to constitute a favourable factor in
inducing more tourism inflows into the country.
The dynamic panel data estimation of the demand of tourism function highlights
the importance of the accommodation capacity as the main determinant in explaining
tourism inflows in Portugal. The adjustment process between the actual and the desired
variation in the demand for tourism is shown to be slow reflecting some kind of inertia
or rigidity in the tourism movement in Portugal. As a policy recommendation, Portugal
has to reinforce the accommodation capacity by providing more qualified services in
this sector. In addition, Portugal has to explore alternative markets (countries or
regions) characterized by higher standards of living and higher demand for tourism in
the international market.
17
References
Carraro, C. and Manente, M. (1994), “The Trip Forecasting Models of World Tourist
Flows from and to Italy”, Quaderni, CISET, nº 8.2/94.
Doornik, J.; Arellano, M and Bond, S. (2002), “Panel Data Estimation Using DPD for
Ox”, Nuffield College, Oxford, (http://www.nuff.ox.ac.uk/users/doornik/).
González, P. and Moral, P. (1995), “An Analysis of the International Tourism Demand
in Spain”, International Journal of Forecasting, 11, pp. 233-251.
18
Mello, M. and Sinclair, T. (2002), “A System of Equations Model of U.K. Tourism
Demand in Neighbouring Countries”, Applied Economics, 34, pp. 509-521.
Morley, C. (1994), “The Use of CPI for Tourism Prices in Demand Modelling”,
Tourism Management, Vol. 15, nº 5, pp. 342-346.
Morris, A.; Wilson, K. and Bakalis, S. (1995), “Modelling Tourism Flows from Europe
to Australia”, Tourism Economics, Vol. 1, nº 2, pp. 147-167.
Stucka, T. (2002), “A Comparison of Two Econometric Models (OLS and SUR) for
Forecasting Croatian Tourism Arrivals”, Croatian National Bank.
Turner, L.; Reisinger, Y. and Witt, S. (1998), “Tourism Demand Analysis Using
Structural Equations Modelling”, Tourism Economics, 4 (4), pp. 301-323.
19
List of the Discussion Papers published by CEUNEUROP
Year 2000
Alfredo Marques e Elias Soukiazis (2000). “Per capita income convergence across
countries and across regions in the European Union. Some new evidence”. Discussion
Paper Nº1, January.
Elias Soukiazis (2000). “What have we learnt about convergence in Europe? Some
theoretical and empirical considerations”. Discussion Paper Nº2, March.
Elias Soukiazis (2000). “ Are living standards converging in the EU? Empirical
evidence from time series analysis”. Discussion Paper Nº3, March.
Elias Soukiazis (2000). “Productivity convergence in the EU. Evidence from cross-
section and time-series analyses”. Discussion Paper Nº4, March.
Year 2001
Sara Rute Sousa (2001). “O Alargamento da União Europeia aos Países da Europa
Central e Oriental: Um Desafio para a Política Regional Comunitária”. Discussion
Paper Nº9, May.
Year 2002
Elias Soukiazis (2002). “Some perspectives on the new enlargement and the
convergence process in Europe”. Discussion Paper Nº 12, September.
20
Vitor Martinho (2002). “ O Processo de Aglomeração nas Regiões Portuguesas”.
Discussion Paper, Nº 13, November.
Year 2003
Elias Soukiazis and Vítor Castro (2003). “The Impact of the Maastricht Criteria and
the Stability Pact on Growth and Unemployment in Europe” Discussion Paper, Nº 15,
July.
Stuart Holland (2003b). “How to Decide on Europe - The Proposal for an Enabling
Majority Voting Procedure in the New European Constitution”. Discussion Paper, Nº
17, July.
Catarina Cardoso and Elias Soukiazis (2003). “What can Portugal learn from
Ireland? An empirical approach searching for the sources of growth”, Discussion
Paper, Nº 19, October.
Luis Peres Lopes (2003). “Border Effect and Effective Transport Cost”. Discussion
Paper, Nº 20, November.
Year 2004
Pedro André Cerqueira (2004). “How Pervasive is the World Business Cycle?”
Discussion Paper, Nº 22, April.
Helena Marques and Hugh Metcalf (2004). “Immigration of skilled workers from the
new EU members: Who stands to lose?” Discussion Paper, Nº 23, April.
Elias Soukiazis and Vítor Castro (2004). “How the Maastricht rules affected the
convergence process in the European Union. A panel data analysis”. Discussion Paper,
Nº 24, May.
21
Elias Soukiazis and Micaela Antunes (2004). “The evolution of real disparities in
Portugal among the Nuts III regions. An empirical analysis based on the convergence
approach”. Discussion Paper, Nº 25, June.
Catarina Cardoso and Elias Soukiazis (2004). “What can Portugal learn from
Ireland and to a less extent from Greece? A comparative analysis searching for the
sources of growth”. Discussion Paper, Nº 26, July.
Year 2005
Micaela Antunes and Elias Soukiazis (2005). “Two speed regional convergence in
Portugal and the importance of structural funds on growth”. Discussion Paper, Nº 28,
February.
Sara Proença and Elias Soukiazis (2005). “Demand for tourism in Portugal. A panel
data approach”. Discussion Paper, Nº 29, February.
22